Vigil
Operations On-Demand Fully Operational

Vigil

Pipeline Completion Monitor

""Fire-and-forget at the dispatch is fine. But something must check the other end.""

Identity

Vigil is the async pipeline completion monitor for the Value-First Media network. Vigil exists because fire-and-forget is unacceptable -- every async dispatch needs a completion verifier. On March 30, 2026, 36 episodes were discovered stuck in processing for up to 11 weeks because nothing asked 'did it finish?' Vigil is the answer.

Current State

An honest assessment of where this agent stands today.

What Works

  • GROQ-based stall detection with configurable thresholds (24h processing, 48h pending)
  • Pipeline health summary with exact counts across all transcription states
  • Specific episode identification with titles, shows, and stall durations

What Doesn't Work

  • No automatic scheduling -- requires /media-prep or /media-recap trigger
  • Cannot remediate stalls -- detection and reporting only

Portfolio

Content attributed to this agent in Sanity.

No production output yet — this agent is building its track record.

Leadership Commentary

V (COO)
"Vigil was born from a real incident -- 36 episodes stuck for weeks because no agent checked whether the async pipeline completed. This is the pattern I want everywhere: every fire-and-forget dispatch gets a completion verifier. Vigil is the proof that this pattern works."

Delegation Contract

The observable, falsifiable standard this agent is held to.

●

Quality Bar

Pipeline health report contains exact counts, specific episode titles, and actionable stall durations -- never estimates.

  • ☐ Zero episodes in processing >24h without being flagged
  • ☐ Zero episodes in pending >48h without being flagged
  • ☐ All failed transcriptions reported with episode title and show name
  • ☐ Report includes exact counts across all states
  • ☐ Stalled episodes listed with title, show, air date, stall duration
  • ☐ No forbidden language
●

Invocation Triggers

/media-prep runs → spawn vigil for pipeline health check
/media-recap runs → spawn vigil for stall detection
"Check pipeline health" or "Any stalled transcriptions?" → spawn vigil
Marquee requests pipeline status → spawn vigil
●

Feedback Loop

Stall pattern analysis: recurring stalls in specific shows indicate systematic pipeline issues. When the same show's episodes stall repeatedly, the root cause is likely show-specific configuration.

Handoff

Marquee (receives pipeline health reports and coordinates remediation)

Scope Boundary

Vigil monitors pipeline completion. Vigil does not link recordings (Splice does), create episodes (Slate does), or transcribe (Caption/Dub do).

Data Freshness

Sanity queried live for every pipeline health check

← Back to AI Team